Be the answer

A buyer used to type a few words into Google and scroll. More and more of them now ask ChatGPT, Claude, Gemini or Perplexity instead, in full sentences, the way they would ask a colleague. The assistant reads, weighs and answers. Usually it names a few companies and stops there.

That answer has no page two. You are named, or for that buyer you do not exist. This year the numbers caught up with the feeling, and they say three things at once: asking has become normal, clicks have become rarer, and the few visitors who do arrive through an assistant come ready to act.

How many people now ask

OpenAI said in February that ChatGPT had 900 million weekly users. By October, TechCrunch put the figure at 1.2 billion. At its I/O conference in May, Google said that AI Overviews, the summaries above its results, reach more than 2.5 billion people a month, that AI Mode, its conversational search, passed one billion monthly users within a year, and that the Gemini app has more than 900 million.

Business buyers have moved just as fast. Forrester, drawing on a survey of nearly 18'000 buyers, reported in January that 94 percent use AI during their buying process. G2 asked 1'076 software buyers in March: 51 percent now start their research with an AI chatbot more often than with Google, up from 29 percent in its 2025 report. 69 percent said they had chosen a different vendor than planned because of what a chatbot recommended, and one in three bought from a vendor they had never heard of before.

That last figure is the one to keep. An assistant does not only confirm the shortlist. It writes it.

Fewer clicks, better visitors

The cost shows up in the clicks. SparkToro, working with clickstream data from Similarweb, found that 68 percent of Google searches in the United States ended without a click in the first four months of 2026, against about 60 percent in 2024. Pew Research Center had already measured the effect of the summaries directly: when one appeared, people clicked a traditional result on 8 percent of visits, against 15 percent without one. Ahrefs found that an AI Overview went with a click rate for the top result about 58 percent lower in December 2025. Each study measures differently. All of them point the same way.

The assistants send little traffic back. In SE Ranking’s sample of 101'574 websites, referrals from ChatGPT reached a record in May 2026 and still made up only about a third of one percent of all visits.

The visits that do come are worth more. Adobe counted 693 percent more traffic from AI tools to American retail sites over the 2025 holidays than a year before, from what it called a modest base. By March 2026 those visitors converted 42 percent better than everyone else, according to Adobe data reported by TechCrunch, where a year earlier they had converted worse. People do their research in the conversation and come to the website to decide.

What the models read

Google has said plainly what its own AI features need, in a guide for site owners updated in July. They build on its ordinary ranking and quality systems. No special schema markup is required, and no AI text files either: Google Search does not use llms.txt, so the file neither helps nor harms a site there. What Google asks for is pages it can crawl and index, and what it calls non commodity content, original and useful rather than a summary of what others have already said.

Other assistants read differently, and some do look for an llms.txt, which is why we still write one. It is a short index and cheap to make. Nobody should sell it as a lever on its own.

The larger problem is more basic. Adobe found that around a third of product pages could not be properly accessed by AI. Text hidden behind scripts, tabs and images is text no model can quote. So the work starts with clear facts in the page itself: what you do, for whom, where, since when, and at what price if the price is public, in the same words on every page. Then pages that survive being cut into pieces: a heading that is the buyer’s question, the answer in the first sentence under it, a date and a name.

Your footprint, not only your site

Your own site is one voice. The models also read what everybody else says about you, and a fact repeated by independent sources weighs more than a fact you state once about yourself. In G2’s survey, a citation from a review site was the signal that most raised buyers’ trust in an AI answer, and 85 percent thought more highly of a vendor when an assistant included it.

For most companies those sources are unglamorous: the commercial register, trade associations, industry directories, comparison pages in the category, review platforms and the local press. Appear where you belong, with the same name, the same description and the same facts as on your own site. Contradictions are the quiet reason a model names someone else.

Where it is going

The answer is turning into a marketplace. OpenAI began showing ads in ChatGPT this year, and in October it announced visual ads in the United States, clearly labelled and, it says, without influence on the answers. Google now lets people in the United States book some hotels inside AI Mode, with the hotel or the booking platform as merchant of record. And assistants are becoming agents: OpenAI’s Dots, launched at the end of September, work in the background with their own browser and ask their user to approve what they do.

For a company this means the shortlist is drawn earlier and by a machine, and the paid place next to the answer will get more crowded. The place inside the answer is still earned with facts. Measurement is catching up as well: Google’s guide now points site owners to a report on generative AI performance in Search Console.

What a company can do

Start with the questions your buyers really ask, the ones from first calls and the contact form, and agree a set per market and per language. Run it every week across ChatGPT, Claude, Gemini, Perplexity and Google, and record who is named, which source was used and who was named instead. Answers vary from run to run, so a single check proves little. A weekly series shows a direction, and out of it comes the number we call share of answer.

Then fix in order: access for the crawlers, facts in the page, pages that can be quoted, the footprint across the internet. Cited does this as a product. We analyse and advise, launch one big update of the website and the footprint together so the models see the change at once, and then measure every week. The same audit runs on every website we ship, crue.ch included. In Studio it sits next to the website, the newsletter and social media, so the facts stay the same everywhere.

What we do not hand to a machine is the substance. The assistants can only quote what a company actually knows and stands for, and that still comes from people: the expertise, the judgement about what to claim, the care to keep it true. Today Cited measures and proposes, and a person approves every change.